1 | // Copyright 2013 The Chromium Authors. All rights reserved. |
2 | // Use of this source code is governed by a BSD-style license that can be |
3 | // found in the LICENSE file. |
4 | |
5 | #ifndef BUTIL_STRINGS_STRING16_H_ |
6 | #define BUTIL_STRINGS_STRING16_H_ |
7 | |
8 | // WHAT: |
9 | // A version of std::basic_string that provides 2-byte characters even when |
10 | // wchar_t is not implemented as a 2-byte type. You can access this class as |
11 | // string16. We also define char16, which string16 is based upon. |
12 | // |
13 | // WHY: |
14 | // On Windows, wchar_t is 2 bytes, and it can conveniently handle UTF-16/UCS-2 |
15 | // data. Plenty of existing code operates on strings encoded as UTF-16. |
16 | // |
17 | // On many other platforms, sizeof(wchar_t) is 4 bytes by default. We can make |
18 | // it 2 bytes by using the GCC flag -fshort-wchar. But then std::wstring fails |
19 | // at run time, because it calls some functions (like wcslen) that come from |
20 | // the system's native C library -- which was built with a 4-byte wchar_t! |
21 | // It's wasteful to use 4-byte wchar_t strings to carry UTF-16 data, and it's |
22 | // entirely improper on those systems where the encoding of wchar_t is defined |
23 | // as UTF-32. |
24 | // |
25 | // Here, we define string16, which is similar to std::wstring but replaces all |
26 | // libc functions with custom, 2-byte-char compatible routines. It is capable |
27 | // of carrying UTF-16-encoded data. |
28 | |
29 | #include <stdio.h> |
30 | #include <string> |
31 | |
32 | #include "butil/base_export.h" |
33 | #include "butil/basictypes.h" |
34 | |
35 | #if defined(WCHAR_T_IS_UTF16) |
36 | |
37 | namespace butil { |
38 | |
39 | typedef wchar_t char16; |
40 | typedef std::wstring string16; |
41 | typedef std::char_traits<wchar_t> string16_char_traits; |
42 | |
43 | } // namespace butil |
44 | |
45 | #elif defined(WCHAR_T_IS_UTF32) |
46 | |
47 | namespace butil { |
48 | |
49 | typedef uint16_t char16; |
50 | |
51 | // char16 versions of the functions required by string16_char_traits; these |
52 | // are based on the wide character functions of similar names ("w" or "wcs" |
53 | // instead of "c16"). |
54 | BUTIL_EXPORT int c16memcmp(const char16* s1, const char16* s2, size_t n); |
55 | BUTIL_EXPORT size_t c16len(const char16* s); |
56 | BUTIL_EXPORT const char16* c16memchr(const char16* s, char16 c, size_t n); |
57 | BUTIL_EXPORT char16* c16memmove(char16* s1, const char16* s2, size_t n); |
58 | BUTIL_EXPORT char16* c16memcpy(char16* s1, const char16* s2, size_t n); |
59 | BUTIL_EXPORT char16* c16memset(char16* s, char16 c, size_t n); |
60 | |
61 | struct string16_char_traits { |
62 | typedef char16 char_type; |
63 | typedef int int_type; |
64 | |
65 | // int_type needs to be able to hold each possible value of char_type, and in |
66 | // addition, the distinct value of eof(). |
67 | COMPILE_ASSERT(sizeof(int_type) > sizeof(char_type), unexpected_type_width); |
68 | |
69 | typedef std::streamoff off_type; |
70 | typedef mbstate_t state_type; |
71 | typedef std::fpos<state_type> pos_type; |
72 | |
73 | static void assign(char_type& c1, const char_type& c2) { |
74 | c1 = c2; |
75 | } |
76 | |
77 | static bool eq(const char_type& c1, const char_type& c2) { |
78 | return c1 == c2; |
79 | } |
80 | static bool lt(const char_type& c1, const char_type& c2) { |
81 | return c1 < c2; |
82 | } |
83 | |
84 | static int compare(const char_type* s1, const char_type* s2, size_t n) { |
85 | return c16memcmp(s1, s2, n); |
86 | } |
87 | |
88 | static size_t length(const char_type* s) { |
89 | return c16len(s); |
90 | } |
91 | |
92 | static const char_type* find(const char_type* s, size_t n, |
93 | const char_type& a) { |
94 | return c16memchr(s, a, n); |
95 | } |
96 | |
97 | static char_type* move(char_type* s1, const char_type* s2, int_type n) { |
98 | return c16memmove(s1, s2, n); |
99 | } |
100 | |
101 | static char_type* copy(char_type* s1, const char_type* s2, size_t n) { |
102 | return c16memcpy(s1, s2, n); |
103 | } |
104 | |
105 | static char_type* assign(char_type* s, size_t n, char_type a) { |
106 | return c16memset(s, a, n); |
107 | } |
108 | |
109 | static int_type not_eof(const int_type& c) { |
110 | return eq_int_type(c, eof()) ? 0 : c; |
111 | } |
112 | |
113 | static char_type to_char_type(const int_type& c) { |
114 | return char_type(c); |
115 | } |
116 | |
117 | static int_type to_int_type(const char_type& c) { |
118 | return int_type(c); |
119 | } |
120 | |
121 | static bool eq_int_type(const int_type& c1, const int_type& c2) { |
122 | return c1 == c2; |
123 | } |
124 | |
125 | static int_type eof() { |
126 | return static_cast<int_type>(EOF); |
127 | } |
128 | }; |
129 | |
130 | typedef std::basic_string<char16, butil::string16_char_traits> string16; |
131 | |
132 | BUTIL_EXPORT extern std::ostream& operator<<(std::ostream& out, |
133 | const string16& str); |
134 | |
135 | // This is required by googletest to print a readable output on test failures. |
136 | BUTIL_EXPORT extern void PrintTo(const string16& str, std::ostream* out); |
137 | |
138 | } // namespace butil |
139 | |
140 | // The string class will be explicitly instantiated only once, in string16.cc. |
141 | // |
142 | // std::basic_string<> in GNU libstdc++ contains a static data member, |
143 | // _S_empty_rep_storage, to represent empty strings. When an operation such |
144 | // as assignment or destruction is performed on a string, causing its existing |
145 | // data member to be invalidated, it must not be freed if this static data |
146 | // member is being used. Otherwise, it counts as an attempt to free static |
147 | // (and not allocated) data, which is a memory error. |
148 | // |
149 | // Generally, due to C++ template magic, _S_empty_rep_storage will be marked |
150 | // as a coalesced symbol, meaning that the linker will combine multiple |
151 | // instances into a single one when generating output. |
152 | // |
153 | // If a string class is used by multiple shared libraries, a problem occurs. |
154 | // Each library will get its own copy of _S_empty_rep_storage. When strings |
155 | // are passed across a library boundary for alteration or destruction, memory |
156 | // errors will result. GNU libstdc++ contains a configuration option, |
157 | // --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which |
158 | // disables the static data member optimization, but it's a good optimization |
159 | // and non-STL code is generally at the mercy of the system's STL |
160 | // configuration. Fully-dynamic strings are not the default for GNU libstdc++ |
161 | // libstdc++ itself or for the libstdc++ installations on the systems we care |
162 | // about, such as Mac OS X and relevant flavors of Linux. |
163 | // |
164 | // See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 . |
165 | // |
166 | // To avoid problems, string classes need to be explicitly instantiated only |
167 | // once, in exactly one library. All other string users see it via an "extern" |
168 | // declaration. This is precisely how GNU libstdc++ handles |
169 | // std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring). |
170 | // |
171 | // This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2), |
172 | // in which the linker does not fully coalesce symbols when dead code |
173 | // stripping is enabled. This bug causes the memory errors described above |
174 | // to occur even when a std::basic_string<> does not cross shared library |
175 | // boundaries, such as in statically-linked executables. |
176 | // |
177 | // TODO(mark): File this bug with Apple and update this note with a bug number. |
178 | |
179 | extern template |
180 | class BUTIL_EXPORT std::basic_string<butil::char16, butil::string16_char_traits>; |
181 | |
182 | #endif // WCHAR_T_IS_UTF32 |
183 | |
184 | #endif // BUTIL_STRINGS_STRING16_H_ |
185 | |